Scikeras Keras Regressor

[Seyed Faraz Eftekhar]

Hi,

I have developed the model in Keras and used Scikeras KerasRegressor for wrapping keras model.

But I was not able to plot the Prediction Error plot.

Any help and recommendation is appreciated :)

[Yellowbrick]

Hello, 

The PredictionError plot requires data in order to plot the residuals against the predicted value; simply passing in the regressor is not sufficient to draw the plot. 

To pass data into the plot, you'll have to call `fit()` and `score()` on the model. However, you obviously don't want to fit the model again (since it probably takes a while to train a DLN). To prevent refitting you can use the `is_fitted` argument as follows:

visualizer = PredictionError(mlp, is_fitted=True) 

visualizer.fit(X_train, y_train)

visualizer.score(X_test, y_test) 

visualizer.show() 

Please note that Yellowbrick was developed for scikit-learn integration specifically. If the scikeras KerasRegressor works - that's great! However, we don't spend a lot of time testing external third-party models. If it doesn't work and you're interested in using Keras with Yellowbrick, we'd appreciate any pull requests into our contrib package to support it! 

Best Regards,

Benjamin Bengfort